The Subjectivity of Respect in Police Traffic Stops: Modeling Community Perspectives in Body-Worn Camera Footage

Open MIND · 2026-02-10

Traffic stops are among the most frequent police-civilian interactions, and body-worn cameras (BWCs) provide a unique record of how these encounters unfold. Respect is a central dimension of these interactions, shaping public trust and perceived legitimacy, yet its interpretation is inherently subjective and shaped by lived experience, rendering community-specific perspectives a critical consideration. Leveraging unprecedented access to Los Angeles Police Department BWC footage, we introduce the first large-scale traffic-stop dataset annotated with respect ratings and free-text rationales from multiple perspectives. By sampling annotators from police-affiliated, justice-system-impacted, and non-affiliated Los Angeles residents, we enable the systematic study of perceptual differences across diverse communities. To this end, we (i) develop a domain-specific evaluation rubric grounded in procedural justice theory, LAPD training materials, and extensive fieldwork; (ii) introduce a rubric-driven preference data construction framework for perspective-consistent alignment; and (iii) propose a perspective-aware modeling framework that predicts personalized respect ratings and generates annotator-specific rationales for both officers and civilian drivers from traffic-stop transcripts. Across all three annotator groups, our approach improves both rating prediction performance and rationale alignment. Our perspective-aware framework enables law enforcement to better understand diverse community expectations, providing a vital tool for building public trust and procedural legitimacy.

Assessments of Credibility in the Social and Behavioral Sciences

MetArXiv (OSF Preprints) · 2026-02-19

Credibility assessment — determining whether research findings are trustworthy or believable — is essential to the research process. One aspect of credibility is repeatability, which includes assessing whether consistent results are obtained when using new data to answer the same question (replicability), when repeating the original analyses with the original data (reproducibility), or when conducting alternative analyses about the same question with the original data (robustness). These features of repeatability differ in the resources required to investigate them, and it is unknown how they relate with one another and with other features of credibility. We investigated relationships among credibility measures in a stratified random sample of claims made across the social and behavioral sciences. Measures of repeatability were modestly correlated with each other (r’s = 0.30, -0.04, -0.23) though the correlation between robustness and reproducibility is likely misestimated because selecting claims for robustness testing was partly contingent on reproducibility success. Replicability and human and machine predictions of replicability were modestly correlated (Median r = 0.23; Range = -0.10 to 0.47). Though estimated with substantial uncertainty in some cases, no discipline showed consistently higher repeatability than other disciplines across measures. For example, Education had the highest replicability estimate (0.63, 95% CI [.32 - .86]) and the lowest reproducibility estimate (0.25, 95% CI [.25, 95% CI [.06 - .38]), whereas Economics had the lowest replicability estimate (0.43, 95% CI [.28 - .64]) and nearly the highest reproducibility estimate (0.70, 95% CI [.56 - .85]). Repeatability measures were modestly and heterogeneously associated with other potential indicators of credibility (Median |r| = 0.08; Range = 0.01 to 0.55). Credibility assessment is multidimensional with substantial opportunity for innovation and validation of its measurement.

Abstraction as a Memory-Efficient Inductive Bias for Continual Learning

arXiv (Cornell University) · 2026-03-17

The real world is non-stationary and infinitely complex, requiring intelligent agents to learn continually without the prohibitive cost of retraining from scratch. While online continual learning offers a framework for this setting, learning new information often interferes with previously acquired knowledge, causes forgetting and degraded generalization. To address this, we propose Abstraction-Augmented Training (AAT), a loss-level modification encouraging models to capture the latent relational structure shared across examples. By jointly optimizing over concrete instances and their abstract representations, AAT introduces a memory-efficient inductive bias that stabilizes learning in strictly online data streams, eliminating the need for a replay buffer. To capture the multi-faceted nature of abstraction, we introduce and evaluate AAT on two benchmarks: a controlled relational dataset where abstraction is realized through entity masking, and a narrative dataset where abstraction is expressed through shared proverbs. Our results show that AAT achieves performance comparable to or exceeding strong experience replay (ER) baselines, despite requiring zero additional memory and only minimal changes to the training objective. This work highlights structural abstraction as a powerful, memory-free alternative to ER.

Assessments of Credibility in the Social and Behavioral Sciences

2026-04-01 · 2 citations

Credibility assessment — determining whether research findings are trustworthy or believable — is essential to the research process. One aspect of credibility is repeatability, which includes assessing whether consistent results are obtained when using new data to answer the same question (replicability), when repeating the original analyses with the original data (reproducibility), or when conducting alternative analyses about the same question with the original data (robustness). These features of repeatability differ in the resources required to investigate them, and it is unknown how they relate with one another and with other features of credibility. We investigated relationships among credibility measures in a stratified random sample of claims made across the social and behavioral sciences. Measures of repeatability were modestly correlated with each other (r’s = 0.30, -0.04, -0.23) though the correlation between robustness and reproducibility is likely misestimated because selecting claims for robustness testing was partly contingent on reproducibility success. Replicability and human and machine predictions of replicability were modestly correlated (Median r = 0.23; Range = -0.10 to 0.47). Though estimated with substantial uncertainty in some cases, no discipline showed consistently higher repeatability than other disciplines across measures. For example, Education had the highest replicability estimate (0.63, 95% CI [.32 - .86]) and the lowest reproducibility estimate (0.25, 95% CI [.25, 95% CI [.06 - .38]), whereas Economics had the lowest replicability estimate (0.43, 95% CI [.28 - .64]) and nearly the highest reproducibility estimate (0.70, 95% CI [.56 - .85]). Repeatability measures were modestly and heterogeneously associated with other potential indicators of credibility (Median |r| = 0.08; Range = 0.01 to 0.55). Credibility assessment is multidimensional with substantial opportunity for innovation and validation of its measurement.

The Subjectivity of Respect in Police Traffic Stops: Modeling Community Perspectives in Body-Worn Camera Footage

ArXiv.org · 2026-02-10

Traffic stops are among the most frequent police-civilian interactions, and body-worn cameras (BWCs) provide a unique record of how these encounters unfold. Respect is a central dimension of these interactions, shaping public trust and perceived legitimacy, yet its interpretation is inherently subjective and shaped by lived experience, rendering community-specific perspectives a critical consideration. Leveraging unprecedented access to Los Angeles Police Department BWC footage, we introduce the first large-scale traffic-stop dataset annotated with respect ratings and free-text rationales from multiple perspectives. By sampling annotators from police-affiliated, justice-system-impacted, and non-affiliated Los Angeles residents, we enable the systematic study of perceptual differences across diverse communities. To this end, we (i) develop a domain-specific evaluation rubric grounded in procedural justice theory, LAPD training materials, and extensive fieldwork; (ii) introduce a rubric-driven preference data construction framework for perspective-consistent alignment; and (iii) propose a perspective-aware modeling framework that predicts personalized respect ratings and generates annotator-specific rationales for both officers and civilian drivers from traffic-stop transcripts. Across all three annotator groups, our approach improves both rating prediction performance and rationale alignment. Our perspective-aware framework enables law enforcement to better understand diverse community expectations, providing a vital tool for building public trust and procedural legitimacy.

The Moral Foundations Reddit Corpus

2026-04-30 · 23 citations

Moral framing and sentiment can affect a variety of online and offline behaviors, including donation, environmental action, political engagement, and protest. Various computational methods in Natural Language Processing (NLP) have been used to detect moral sentiment from textual data, but achieving strong performance in such subjective tasks requires large, hand-annotated datasets. Previous corpora annotated for moral sentiment have proven valuable, and have generated new insights both within NLP and across the social sciences, but have been limited to Twitter. To facilitate improving our understanding of the role of moral rhetoric, we present the Moral Foundations Reddit Corpus, a collection of 16,123 English Reddit comments that have been curated from 12 distinct subreddits, hand-annotated by at least three trained annotators for 8 categories of moral sentiment (i.e., Care, Proportionality, Equality, Purity, Authority, Loyalty, Thin Morality, Implicit/Explicit Morality) based on the updated Moral Foundations Theory (MFT) framework. We evaluate baselines using large language models (Llama3-8B, Ministral-8B) in zero-shot, few-shot, and PEFT (Parameter-Efficient Fine-Tuning) settings, comparing their performance to fine-tuned encoder-only models like BERT (Bidirectional Encoder Representations from Transformers). The results show that LLMs continue to lag behind fine-tuned encoders on this subjective task, underscoring the ongoing need for human-annotated moral corpora for AI alignment evaluation. Keywords: moral sentiment annotation, moral values, moral foundations theory, multi-label text classification, large language models, benchmark dataset, evaluation and alignment resource

Abstraction as a Memory-Efficient Inductive Bias for Continual Learning

ArXiv.org · 2026-03-17

The real world is non-stationary and infinitely complex, requiring intelligent agents to learn continually without the prohibitive cost of retraining from scratch. While online continual learning offers a framework for this setting, learning new information often interferes with previously acquired knowledge, causes forgetting and degraded generalization. To address this, we propose Abstraction-Augmented Training (AAT), a loss-level modification encouraging models to capture the latent relational structure shared across examples. By jointly optimizing over concrete instances and their abstract representations, AAT introduces a memory-efficient inductive bias that stabilizes learning in strictly online data streams, eliminating the need for a replay buffer. To capture the multi-faceted nature of abstraction, we introduce and evaluate AAT on two benchmarks: a controlled relational dataset where abstraction is realized through entity masking, and a narrative dataset where abstraction is expressed through shared proverbs. Our results show that AAT achieves performance comparable to or exceeding strong experience replay (ER) baselines, despite requiring zero additional memory and only minimal changes to the training objective. This work highlights structural abstraction as a powerful, memory-free alternative to ER.

Reasoning on a Spectrum: Aligning LLMs to System 1 and System 2 Thinking

arXiv (Cornell University) · 2025-02-18

Large Language Models (LLMs) exhibit impressive reasoning abilities, yet their reliance on structured step-by-step processing reveals a critical limitation. In contrast, human cognition fluidly adapts between intuitive, heuristic (System 1) and analytical, deliberative (System 2) reasoning depending on the context. This difference between human cognitive flexibility and LLMs' reliance on a single reasoning style raises a critical question: while human fast heuristic reasoning evolved for its efficiency and adaptability, is a uniform reasoning approach truly optimal for LLMs, or does its inflexibility make them brittle and unreliable when faced with tasks demanding more agile, intuitive responses? To answer these questions, we explicitly align LLMs to these reasoning styles by curating a dataset with valid System 1 and System 2 answers, and evaluate their performance across reasoning benchmarks. Our results reveal an accuracy-efficiency trade-off: System 2-aligned models excel in arithmetic and symbolic reasoning, while System 1-aligned models perform better in commonsense reasoning tasks. To analyze the reasoning spectrum, we interpolated between the two extremes by varying the proportion of alignment data, which resulted in a monotonic change in accuracy. A mechanistic analysis of model responses shows that System 1 models employ more definitive outputs, whereas System 2 models demonstrate greater uncertainty. Building on these findings, we further combine System 1- and System 2-aligned models based on the entropy of their generations, without additional training, and obtain a dynamic model that outperforms across nearly all benchmarks. This work challenges the assumption that step-by-step reasoning is always optimal and highlights the need for adapting reasoning strategies based on task demands.

Targeting audiences’ moral values shapes misinformation sharing.

Journal of Experimental Psychology General · 2025-01-13 · 5 citations

= 20,235; 809,414 tweets) that explores how aligning the moral values of message senders with misinformation content influences its dissemination in the context of COVID-19 vaccination misinformation. First, we investigate how aligning messages' moral framing with participants' moral values impacts participants' intentions to share true and false news headlines and whether this effect is driven by a lack of analytical thinking. Our results show that framing a post such that it aligns with audiences' moral values leads to increased sharing intentions, independent of headline familiarity, and participants' political ideology but find no effect of analytical thinking. Furthermore, we find that moral alignment facilitates sharing misinformation more so than true information. Next, we use natural language processing to determine messages' moral framing and senders' political ideology. We find that an alignment of moral framing and ideology facilitates the spread of misinformation. Our findings suggest that (a) targeting audiences' core values can be used to influence the dissemination of (mis)information on social media platforms; (b) partisan divides in misinformation sharing can be, at least partially, explained through alignment between audiences' underlying moral values and moral framing that often accompanies content shared online; and (c) this effect is driven by motivational factors. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Author response for "Syntactic and Semantic Gender Biases in the Language on Children’s Television: Evidence From a Corpus of 98 Shows From 1960 to 2018"

2025-05-09